Eddy/topup on denoised data

1

Dear MRtrix experts,

I’m experiencing some problems when running topup/eddy on denoised DWI data.

I have two DWI runs with reversed PE direction, each run has 30 directions with b=1000 and a few b=0 images.

This is what I have done:

1. denoise data

dwidenoise dwi.nii.gz dwi_denoised.nii.gz
dwidenoise dwi_revPE.nii.gz dwi_revPE_denoised.nii.gz

2. degibbs denoised data

mrdegibbs dwi_denoised.nii.gz dwi_denoised_unringed.nii.gz
mrdegibbs dwi_revPE_denoised.nii.gz dwi_revPE_denoised_unringed.nii.gz

3. create b0_revB0 image

fslroi dwi_denoised_unringed.nii.gz dwi_denoised_unringed_b0.nii.gz 0 1
fslroi dwi_revPE_denoised_unringed.nii.gz dwi_revPE_denoised_unringed_b0.nii.gz 0 1
fslmerge -t denoised_unringed_b0_revb0 dwi_denoised_unringed_b0.nii.gz dwi_revPE_denoised_unringed_b0.nii.gz

4. run topup

topup --imain=denoised_unringed_b0_revb0.nii.gz --datain=acqparams.txt --config=b02b0.cnf --out=denoised_unringed_b0_revb0_topup

5. create brain mask

applytopup --imain=dwi_denoised_unringed.nii.gz --inindex=1 --datain=acqparams.txt --topup=denoised_unringed_b0_revb0_topup --method=jac --out=dwi_denoised_unringed_cor
fslroi dwi_denoised_unringed_cor.nii.gz dwi_denoised_unringed_cor_b0 0 1
bet dwi_denoised_unringed_cor_b0.nii.gz dwi_denoised_unringed_cor_b0_brain -m -f 0.2

6. run eddy with topup

eddy --imain=dwi_denoised_unringed.nii.gz --mask=dwi_denoised_unringed_cor_b0_brain_mask.nii.gz --index=index.txt --acqp=acqparams.txt --bvecs=dwi.bvec --bvals=dwi.bval --topup=denoised_unringed_b0_revb0_topup --out=dwi_denoised_unringed_eddy

This is what the input DWI (without any preprocessing) looks like (dwi.nii.gz):
dwi

This is the output from topup/eddy (dwi_denoised_unringed_eddy.nii.gz):
dwi_denoised_unringed_eddy

It looks to me that the movement artefacts after running topup+eddy are similar to the raw data.

If I run the same steps, but without denoising and unriniging the data (so starting from step 3 above), the eddy output looks much better with less residual movement:
dwi_eddy

I have also tried it without topup and the problem is the same.

Has anyone experienced a similar issue? Is it possible that applying eddy on denoised data leads to something weird for some datasets?

Thanks a lot in advance for your help!
Kind regards,
Julia

2

Dear Julia,

I had the same problem some years ago, albeit with a b = 3000 dataset. I reached out to the FSL mailing list for help. Here is a part of a discussion with Jesper Anderson (he is one of the authors of eddy):

Yes, I don’t recommend using denoising prior to eddy. I have seen very poor results on denoised data. It tends to mean that one ends up with unrealistic estimates of hyperparameters for the Gaussian Process, making “everything” look like valid signal variance and nothing look like noise. It would also interfere with eddy_qc and lead to inflated estimates of CNR.
I would also have thought that any denoising would work much better after eddy, where for example outliers have been detected and corrected.

That should answer your question :slight_smile:

Best regards

Samuel

3

Hi Julia,
I posted this question on the qsiprep github (Eddy on denoised data. Is this optimal? · Issue #948 · PennLINC/qsiprep · GitHub). Please check out the link to their paper, which indicates that EDDY performed better when denoising was done first.

Cheers
William